Digital Signal Processing Chapter 10. Fourier Analysis of Discrete- Time Signals and Systems CHI. CES Engineering. Prof. Yasser Mostafa Kadah

Transcription

1 Digital Signal Processing Chapter 10 Fourier Analysis of Discrete- Time Signals and Systems Prof. Yasser Mostafa Kadah CHI CES Engineering

2 Discrete-Time Fourier Transform Sampled time domain signal has a Fourier transform that is periodic Periodic time domain signal has a Fourier transform that is sampled Periodic and sampled time domain signal has a Fourier transform that is both periodic and sampled Forward DTFT Inverse DTFT

3 Notes on DTFT DTFT is just one period of the continuous Fourier transform of the sampled signal Z-transform is related to the DTFT as follows: DTFT is the values on the unit circle in z-transform Unit circle must be included in the ROC Problem: the sampled signal is still infinite in length Impossible to obtain using computers Consider cases when the signal has compact support

4 Discrete Fourier Transform (DFT) Defined as the Fourier transform of a signal that is both discrete and periodic Fourier transform will also be discrete and periodic Can assume periodicity if we have a finite duration of a signal: better than zero assumption since it allows reducing the frequency domain to a sampled function

5 DFT Formula Given a periodic signal x[n] of period N, its DFT is given by, Its inverse is given by, Both X[k] and x[n] are periodic of the same period N.

6 Computation of DFT Using FFT A very efficient computation of the DFT is done by means of the FFT algorithm, which takes advantage of some special characteristics of the DFT DFT requires O(N2) computations FFT requires O(N log 2 N) computations Both compute the same thing It should be understood that the FFT is not another transformation but an algorithm to efficiently compute DFTs. Always remember that the signal and its DFT are both periodic

7 Improving Resolution in DFT When the signal x[n] is periodic of period N, the DFT values are normalized Fourier series coefficients of x[n] that only exist for the N harmonic frequencies {2 k/n}, as no frequency components exist for any other frequencies The number of possible frequencies depend on the length L chosen to compute its DFT. Frequencies at which we compute the DFT can be seen as frequencies around the unit circle in the z-plane We can improve the frequency resolution of its DFT by increasing the number of samples in the signal without distorting the signal by padding the signal with zeros Increase samples of the units circle

8 Phase in DFT Output of DFT is complex numbers Phase is just as important as magnitude in DFT Time shift amounts to linear phase Phase wrapping occurs for large shifts Linear phase can be unwrapped to its correct form using the unwrap function of Matlab N= 32; x= zeros(n,1); x(11:15)= [ ]; X= fftshift(fft(x)); subplot(4,1,1); stem(1:n,x); title('signal') subplot(4,1,2); plot(abs(x)); title('dft Magnitude') subplot(4,1,3); plot(angle(x)); title('computed Phase') subplot(4,1,4); plot(unwrap(angle(x))); title('unwrapped Phase')

9 Phase in DFT: Examples Less Time Shift More Time Shift Slower linear phase Steeper linear phase

10 Zero-Padding Example Matlab Code N= 32; N2= 128; n=0:n-1; n2= 0:N2-1; x= zeros(n,1); x(1:16)= 1; X= fftshift(fft(x)); X2= fftshift(fft(x,n2)); subplot(3,1,1) stem(n,x) subplot(3,1,2) stem(n,abs(x)) subplot(3,1,3) stem(n2,abs(x2)) Original Signal 32-Point FFT 128-Point FFT No zero-padding With zero-padding

11 Linear vs. Circular Convolution Convolution in continuous and discrete time aperiodic signals is called linear convolution Convolution can be calculated as a multiplication in the frequency domain Convolution property of the Fourier transform Problem: when using DFT to compute convolution of aperiodic signals, the periodicity assumption results in errors in computing the convolution Signal outside the N points used is not zero: periodic extension of the N points disturb the computation What we compute is called circular convolution

12 Linear vs. Circular Convolution Given x[n] and h[n] of lengths M and K, the linear convolution sum y[n] of length N=(M+K-1) can be found by following these three steps: Compute DFTs X[k] and H[k] of length L N for x[n] and h[n] using zero-padding Multiply them to get Y[k]= X[k]H[k]. Find the inverse DFT of Y[k] of length L to obtain y[n]

13 Linear vs. Circular Convolution: Example N1= 16; N2= 16; N= 32; % N>N1+N2-1 x1= zeros(n1,1); x2= zeros(n2,1); x1(1:4)= 1; x2(:)= 1; X1= fft(x1,n1); X2= fft(x2,n1); X_circ= X1.*X2; x_circ= ifft(x_circ); X1= fft(x1,n); X2= fft(x2,n); X_lin= X1.*X2; x_lin= ifft(x_lin); figure(1); subplot(4,1,1); stem(1:n1,x1) subplot(4,1,2); stem(1:n2,x2) subplot(4,1,3); stem(1:n1, abs(x_circ)) title('circular Convolution'); subplot(4,1,4); stem(1:n,abs(x_lin)) title('circular Convolution');

14 Fourier Transformations Chart Time Signal Classification Continuous Discrete Continuous Continuous Fourier Transform (CFT) Discrete-Time Fourier Transform (DTFT) Frequency Discrete Discrete Fourier Series (DFS) Discrete Fourier Transform (DFT) Most General Form: others are special cases Only one that can be done on a computer

15 Problem Assignments Problems: 10.12, 10.26, 10.27, Partial Solutions available from the student section of the textbook web site